When do the bootstrap servers for external listeners change?

[Otto45]

We are running strimzi kafka in an AWS EKS cluster, so when we have EKS perform a k8s or ami upgrade for our managed node group, it will spin up new EC2 instances and drain the old ones. Obviously when this occurs, our bootstrap server list for our external listener will change, as the new EC2 instances for the node group will have different private IPs. We have noticed that our EC2 instances in our node group actually have 3 different private IPs each (currently we're unsure why this is). Would there be any reason the strimzi operator would re-evaluate the IPs used as the bootstrap servers for a listener and change them during a reconciliation run? We have had this happen to us, but we're unsure if something in AWS was tweaked which would cause this, so I want to rule out the strimzi operator behaving in this manner.

Here is the config for our external listener:

- name: externtls
  port: 9096
  type: nodeport
  tls: true
  authentication:
    type: tls
  configuration:
    preferredNodePortAddressType: InternalIP

[scholzj]

I'm not sure I follow the question / issue. You use nodeport listener. So the brokers will be exposed on the nodes based on their address - in your case based on their internal IP. So if the node where the broker is running changes (its internal IP changes), the brokers will get new advertised addresses.

I'm not sure what exactly you call bootstrap servers. Bootstrap servers are not something configured in the Kafka cluster per se. It can be basically any address that routes to the brokers. So you can for example set up some DNS pointing to all your nodes or something like that if you want.

[Otto45]

I'm using the normal definition of "bootstrap server" used in kafka clients, so it's what you said. Our issue is that when we call the following command to list the bootstrap servers for our external listener, the list changes periodically:
kubectl get kafka kafka-datapipeline-main -o jsonpath='{.status.listeners[?(@.name=="externtls")].bootstrapServers}
I stated that this will definitely happen when we perform an upgrade on our node group in AWS of some kind, as we are getting new servers and hence new IPs. What I don't understand is why this seemingly happens when we have not made any changes to our node group. Each node in our node group has 3 private IPs assigned by AWS, so I was curious if it's possible the strimzi operator periodically changes the IPs used as the addresses for the brokers. We have one broker per EC2 instance currently.

[scholzj]

Well, that address is created based on the advertised addresses of the individual brokers. So that will change when the brokers are scheduled to different nodes. There are not many other ways how to create an address we know works for sure.

But as I said, you can use other ways to bootstrap the clients as well ... you know your infrastructure better and unlike the operator, you do not need to deal with other environments. So if you have some stable address you can use, you can just go and use it.